Earnings Smoothing Activities of Firms to Manage
Credit Ratings*
Boochun Jung
University of Hawaii at Manoa
School of Accountancy
Shidler College of Business
2404 Maile Way
Honolulu, HI 96822
Boochun@hawaii.edu
Naomi Soderstrom
University of Colorado at Boulder
Leeds School of Business
419 UCB
Boulder, CO 80309-0419
Naomi.Soderstrom@colorado.edu
Sunny Yang
University of Texas at Austin
Red McCombs School of Business
Department of Accounting, CBA 4M.250
1 University Station B6400
Austin, TX 78712
Sunny.Yang@mccombs.utexas.edu
First Draft: April 2007
Current Draft: April 2009
*We thank comments from Sanjai Bhagat, Michael Clement, Katherine Gunny, Ross Jennings,
Bjorn Jorgensen, Ghon Rhee, Steve Rock, Laura Starks, Han Yi, workshop participants at Seoul
National University, the Capital Market Reading Group at UT-Austin Accounting Department,
the University of Arizona, the 2007 Accounting-Finance Mini-conference at UT-Austin, and the
2008 Korean Accounting Association annual meeting.
Earnings Smoothing Activities of Firms to Manage
Credit Ratings
Abstract
This paper explores the effect of incentives to maintain or improve credit ratings on corporate
financial reporting. Credit ratings have significant cost implications for companies because they
are directly related to companies’ borrowing cost, and ratings changes are reflected in both stock
returns and bond yields. Managers therefore have incentives to maintain or improve their
companies’ credit ratings. Both managers and rating agencies view earnings smoothness as an
important input into the rating process. We argue that firms have incentives to manage credit
ratings through smoothing earnings and as a result, the degree of earnings smoothing activity is
larger for those with greater incentives. We hypothesize that firms within broad rating categories
(e.g., AA) have differential incentives to smooth earnings, depending on whether their rating is at
the top or bottom notch of the rating category (e.g., AA+ or AA-, respectively) versus in the
middle of the rating category (e.g., AA). This increased incentive stems from a higher probability
of being upgraded (downgraded) into a higher (lower) broad rating category for firms in the
outer notches. Our empirical evidence is consistent with increased earnings smoothing for these
firms. We further hypothesize and find that those firms in the outer notches who rely more
heavily on debt financing smooth earnings to a larger extent than firms who do not rely as much
on debt financing. Finally, we examine whether smoothing behavior has an impact on
subsequent credit ratings. We find that increased earnings smoothness reduces (raises) the
likelihood of a rating downgrade (upgrade) in the following period.
JEL Classification: M41
Keywords: Earnings smoothing; Credit ratings
Data Availability: All data used in this study are publicly available from the sources identified
in the text.
Earnings Smoothing Activities of Firms to Manage Credit Ratings
1. Introduction
Graham et al. (2005) provide evidence that managers greatly value smooth earnings. In
their survey of 401 executives, 97 percent of managers indicate that they prefer a smooth
earnings path even when cash flow volatility is constant. Despite the emphasis of managers on
earnings smoothing and the vast literature on why managers prefer smooth earnings, 1 to our
knowledge, the literature has not provided a link between the importance that managers place on
smooth earnings and corporate reporting behavior. We investigate whether the strength of
incentives to manage perceived debt-paying ability, and thus credit ratings (Trueman and Titman
1988), is related to mangers’ discretionary earnings smoothing activity. We also investigate the
effectiveness of this strategy on rating agencies’ assessment of credit ratings.
We focus on the incentive to improve or maintain credit ratings because ratings play a
key role in corporate financing and investment decisions (Blume et al. 1998) and rating changes
have significant cost implications for companies (e.g., Holthausen and Leftwich 1986;
Ederington and Goh 1998; Dichev and Piotroski 2001). More importantly, despite the
commonly-recognized importance of credit ratings to firms, the effect of the incentive to
maintain or improve credit ratings on managers’ discretionary behavior is largely unknown.
A necessary condition for incentives to manage credit ratings affecting earnings
smoothing behavior is that earnings smoothness must be viewed as a key rating factor by both
managers and rating agencies. Graham et al. (2005) confirm managers’ perceived importance of
smooth earnings. Approximately 42 percent of the managers in their survey believe that
1
Various incentives are examined in the literature. For example, managers smooth earnings to improve
predictability of future earnings (Barnea et al. 1975; Tucker and Zarowin 2006), to improve long-term stock price
(Goel and Thakor 2003), to reduce perceived bankruptcy risk (Trueman and Titman 1988), to reduce the probability
of being fired (Fudenberg and Tirole 1995), and to protect managers’ private control benefits (Leuz et al. 2003).
1
smoother earnings enable them to preserve current credit ratings or achieve better credit ratings.
Rating agencies’ description of their rating process and rating decisions reveal that they consider
earnings smoothness to be a key rating factor. For example, Moody’s lists earnings volatility as
one of the key rating factors in its rating methodology (Moody’s Rating Methodology, June
2006). In explaining its rating decisions for Freddie Mac (RatingsDirect report, December 29,
2006) and Sony (Business Week, November 22, 2004), S&P cited earnings volatility as a key
rating issue. Empirically, Gu and Zhao (2006) document a significant relation between credit
ratings and earnings volatility, with earnings volatility dominating cash flow volatility in
explaining credit ratings.
Both the perceived importance of earnings smoothness by managers and the reliance of
rating agencies on earnings smoothness in their rating decisions increase managers’ incentives to
smooth earnings. However, there are several factors that may impact the strength of these
incentives. First, earnings smoothing is costly because it consumes mangers’ time and effort.
Second, firms can lose credibility among stakeholders when their discretionary behavior is
detected (Goel and Thakor 2003). Thus, it is not always optimal to smooth earnings (Gonedes
1972). In addition, credit rating agencies have substantial expertise and have access to a broad
set of information, including private information, so they are in a good position to detect bond
issuers’ earnings smoothing activities. Rating agencies’ ability to detect discretionary earnings
smoothing could discourage managers from undertaking such activity (Jorion et al. 2005).2
Finally, even if rating agencies are capable of completely detecting earnings smoothing,
they may lack incentive to do so. On one hand, rating agencies must retain their reputations to
continue attracting revenues (Smith and Walter 2001). This provides incentives for the rating
2
However, in the recent mortgage market turmoil, the superior expertise of rating agencies has been questioned. See,
for example, “Failing grades? Why regulators fear credit rating agencies may be out of their depth,” Financial Times,
17 May 2007.
2
agencies to detect and adjust for bond issuers’ opportunistic activities. In an article discussing
rating agencies and the current sub-prime mortgage crisis, Brian Clarkson, former president and
chief operating officer for Moody’s, commented on the general role of rating agencies:
“We join others who have emphasized the crucial role trust and transparency play in the
capital markets and recognize the function rating agencies play in supporting those
goals…We are committed to doing the best job possible for investors by providing
informed and unbiased opinions about the credit risk of securities we rate” (Financial
Times, 18 September 2007, p. 42).
On the other hand, rating agencies’ incentives to discourage opportunistic behavior may
be compromised by conflicts of interest resulting from their dependence on rating fees paid by
issuers and pressure from issuers, investment banks, and commercial banks (Covitz and Harrison
2003; Economist 2005).3 In summary, given the cost of smoothing earnings for management,
rating agencies’ ability (or lack thereof) to detect earnings smoothing, and their opposing
incentives to detect earnings smoothing behaviors due to reputational concerns versus
dependence on rating fees, whether managers smooth earnings to manage credit ratings becomes
an empirical question.
Drawing on the literature (Kisgen 2006; Shah 2007), we examine a setting where we can
identify managers with stronger incentives to preserve current credit ratings or achieve higher
credit ratings. These are firms whose credit ratings are in the upper or lower end of each broad
rating category (i.e., having either a “+” or a “-”appended to their broad rating category). We
refer to these firms as being near a rating change to another broad category.4 Given that most
rating changes are by one “notch,” firms with a + or – notch rating are more likely to shift among
broad categories than are firms in the middle notch (i.e., be upgraded to the next higher broad
3
See the October 8, 2002 Report of the Staff to the Senate Committee on Governmental Affairs, Financial Oversight
of Enron: the SEC and Private-Sector Watchdogs, page 113. Also see “Rating agencies hit by subprime probe,”
Financial Times, 16 August 2007.
4
Throughout the paper, we interchangeably use the phrases: being near a rating change, being near a downgrade or
upgrade to another broad rating category, and being in the top or bottom of a broad category.
3
category in the case of a + notch, or be downgraded to the next lower broad category in the case
of a – notch). The stronger incentives arise from regulators’ general focus on broad ratings
rather than “notch” ratings (Kisgen 2006), a greater yield difference between two adjacent
ratings that cross broad rating categories than between two ratings within the same broad
category (Shah 2007), and a stronger market response to one-notch rating revisions that are
across two adjacent broad ratings than to revisions within the same broad rating category (see the
Appendix). Thus, to test whether managers smooth earnings to manage credit ratings, we
examine whether managers of firms with a + or – notch rating are more likely to smooth earnings
than managers of other firms.
We find that firms close to a rating change smooth earnings to a larger extent than do
other firms within the same broad rating category. In addition, among firms near a rating change,
those with higher financing needs smooth earnings to a larger extent than others. Finally, we find
a negative (positive) relation between an increase in earnings smoothness and the likelihood of a
subsequent rating downgrade (upgrade). A wide range of sensitivity analyses do not change our
main inferences. For example, the results are robust to three different measures of earnings
smoothing activities and use of absolute value of discretionary accruals as a proxy for
discretionary earnings management. The results also generally hold among both firms that meet
earnings thresholds (such as analyst forecast, prior year earnings, or zero earnings) and those that
miss earnings thresholds, suggesting that the short term incentives to meet earnings thresholds do
not entirely drive or subsume the smoothing activity related to debt rating changes.
Our analysis focuses on longer term earnings smoothing rather than on shorter term
accruals management (e.g., signed discretionary accruals) for two reasons. First, credit ratings
are meant to discriminate relative long-term risk (Moody’s 2002). To evaluate the credit-
4
worthiness of issuers, credit rating agencies consider both the short-term earnings improvement
and its persistence. Use of discretionary accruals management to improve earnings increases
short-term earnings, but these increases are unlikely to persist due to later accrual reversals. Such
reversals increase the volatility of earnings and reduce the predictability of future earnings.
Second, we examine issuers’ general earnings management tendency that does not depend on
specific events, such as when firms are issuing new debt or face a rating change. Thus, even if
some sample firms have the tendency to manage earnings upward upon those events, the
incentive to smooth earnings would dominate in our general setting.5
This study contributes to the literature in several ways. First, prior studies primarily
document opportunistic reporting behaviors associated with equity market incentives. However,
firms that participate in the public debt market also have incentives to improve perceptions of
their financial health and, thus, may also engage in opportunistic behaviors such as earnings
management. By focusing on incentives related to managing perceptions of a single group of
well-informed users (i.e., bond rating agencies), we avoid confounding effects in more traditional
equity market settings from other incentives and investors’ differential abilities and investment
goals (e.g., speculation versus long-term investment). Second, prior literature either examines
incentives to smooth earnings or the consequence of this behavior. We extend the literature by
investigating variation of managers’ earnings smoothing behavior associated with the strength of
a specific incentive. Finally, prior literature on debt-related earnings management focuses on
firms that move close to default on their debt covenants (Watts and Zimmerman 1986). We
5
In unreported analyses we find that firms with stronger incentives to maintain or improve credit ratings are not
more likely to report earnings that surpass earnings thresholds or have a higher magnitude of signed discretionary
accruals. This is consistent with Graham et al. (2005) in that managers appear less interested in meeting or beating
earnings benchmarks to improve credit ratings. If issuers manage accruals either upward or downward to smooth
earnings, however, the absolute value of discretionary accruals should be larger for firms with stronger incentives.
Our results in section 5.1 support this.
5
employ a sample of firms within a broad spectrum of financial health, ranging from investmentgrade to speculative-grade debt and a setting where management’s differential incentives to
manage earnings can be identified. We are thus able to provide broader inferences about debtrelated earnings management.
The remainder of the paper is organized as follows. The next section develops hypotheses
relating credit ratings to earnings smoothing incentives. Section 3 outlines the research designs.
Section 4 describes the sample, presents descriptive statistics, and reports the results of our main
tests. Section 5 includes sensitivity analyses, and section 6 concludes.
2. Literature review and hypothesis development
2.1. Background on credit ratings and related incentives
Credit rating agencies such as Standard & Poor’s and Moody’s evaluate credit risk and
assign credit ratings to issues, issuers, or both. Credit risk is defined as the possibility that a bond
issuer will default by failing to make principal and interest payments under the bond’s terms. 6 A
credit rating is basically an opinion provided by a rating agency as to credit risk. The primary
credit rating category is most commonly denoted with letter ratings. For example, S&P uses ten
“broad” categories (i.e., AAA, AA, A, BBB, BB, B, CCC, CC, C, D). S&P divides each broad
category from AA to CCC into three subcategories, or “notches” (+/none/-).7 If a rating agency
announces a downgrade (upgrade), it means that the debt rating agency believes that the issuer’s
or the issue’s credit quality has decreased (increased) and the probability of default has increased
(decreased).
Ratings have significant cost implications for companies. Not only do ratings impact the
cost of future borrowing, but numerous studies (e.g., Holthausen and Leftwich 1986; Ederington
The definition we use follows S&P’s. Moody’s defines credit risk as a relative expected loss rate, which is the
product of expected default rates and expected loss-severity rates in the case of default.
7
Moody’s also uses an alphabetic system, but augments the letter ratings with numbers instead of +/-.
6
6
and Goh 1998; Dichev and Piotroski 2001) show that rating changes significantly affect both
stock and bond valuations. For example, Dichev and Piotroski (2001) document a three-day
market response to credit rating downgrades of -1.97 percent, suggesting that the economic costs
associated with rating downgrades are substantial.
The significant stock and bond market response to rating changes creates incentives for
bond issuers to improve or maintain their credit ratings. Recent studies suggest that incentives
vary among bond issuers, and those with “+” or “-” notch ratings have stronger incentives than
other firms in the same broad rating category (Kisgen 2006; Shah 2007). The incentive variation
arises from variations in regulatory costs, bond yield, and stock market response to rating
changes that are of the same magnitude, but which result in movement across versus within
broad rating categories.
Kisgen (2006) argues that regulators generally do not focus on notch ratings, so firms are
more concerned about changes in their broad rating category.8 For example, A and A- are
generally considered as similar from a regulatory perspective. If a firm with A- is downgraded by
one level to BBB+, it will be pooled with all firms within the broad BBB rating, including
BBB+, BBB, and BBB-, rather than being viewed as just one notch below A-. The firm will thus
incur higher regulatory costs than if it were an A firm moving down to A-. Kisgen finds that
firms near rating changes to an adjacent broad rating category are less likely to issue debt than
equity because a significant debt offering might trigger a downgrade or prohibit an upgrade.
Shah (2007) documents a greater change in yield between two adjacent ratings across broad
Regulators’ focus on broad ratings is also described in a rating scoring system to compute capital requirements for
investments made by insurance companies (Kisgen 2006). Investments in firms rated A- or above are assigned a
value of 1. Investment in firms with a rating in the broad category of BBB, BB, B, C, and D are assigned a value of
2,3,4,5, and 6, respectively. Thus defined, a rating scoring system ignores notch ratings and focuses on only broad
ratings.
8
7
rating categories than between two ratings within the same broad category.9 Shah (2007) reports
that compared to other firms, firms near a rating downgrade tend to spend less on investments.
Thus, according to Kisgen (2006) and Shah (2007), credit ratings are important to firms, and
managers’ actions appear to be consistent with their incentives to maintain or improve current
ratings.
In the appendix, we provide further evidence of firms near a rating change having more
powerful incentives to influence credit ratings. We document a stronger market response to onenotch rating revisions across two adjacent broad ratings than to revisions within the same broad
rating category. Untabulated analysis also shows that 64 percent of credit rating revisions
between April 1995 and December 2006 are by one notch (73 percent for investment grade and
55 percent for speculative grade). Thus, compared to firms in the middle notch, firms with a plus
or minus notch rating are more likely to experience ratings changes that move them to an
adjacent broad rating category and have greater incentives to increase (decrease) the likelihood
of being upgraded (downgraded).
2.2. Literature on credit ratings and earnings smoothness
Financial performance is an important determinant of credit ratings. Although other
factors, such as industry risk, capital structure, and management efficiency, affect credit ratings,
these factors appear to be less salient than financial performance. For example, Goh and
Ederington (1993) find that firm value is more significantly affected by credit rating changes
triggered by financial performance than by other factors. They provide evidence that the stock
For example, Shah (2007) compares the mean yield difference between ratings having a “+” notch and two
adjacent rating groups. The first group has a “-” notch rating and is in a different broad rating category and the
second group has a middle notch and is within the same broad rating category. The difference in yield is 47.7 basis
point and 20.5 basis points for the first and the second group, respectively. Thus, the change in yield is 129 percent
higher for a movement across broad rating categories than within a single broad rating category. Comparison of
ratings having a “-” notch with two neighboring rating groups yields similar conclusion.
9
8
market reacts negatively to only those downgrades claimed by rating agencies as being
associated with deterioration in a firm’s financial measures.
We focus on managing earnings smoothness as a means to manage credit ratings not only
because earnings smoothness is an important input into the rating methodology, but also because
Graham et al. (2005) find that managers’ preference for a smooth earnings path is very strong.
For example, managers claim that they prefer smoother earnings even if it means sacrificing
long-term firm value, since a more volatile earnings path can result in an increase in the cost of
debt and equity (Graham et al. 2005).10
Managers can reduce the probability of default risk as perceived by both investors and
rating agencies by smoothing earnings. In an analytical study, Trueman and Titman (1988) show
that reducing volatility of the underlying earnings process can affect the perceived probability of
bankruptcy risk. Beaver et al. (1970) show that earnings smoothness impacts perceived firm risk.
Francis et al. (2005) show that debt markets price information risk associated with earnings
volatility. The market also appears to associate earnings volatility with debt-related costs. Collins
et al. (1981) and Lys (1984) find negative returns at announcements of accounting rule changes
that are predicted to increase earnings volatility, with more negative returns for firms with
stricter debt constraints.
2.3. Hypothesis development
As discussed earlier, rating changes have significant cost and benefit implications for
bond issuers in both the debt and equity markets. This provides incentives for bond issuers to
10
To date, no consensus exists as to whether earnings smoothing reduces the cost of equity. On one hand, smoother
earnings enhance earnings predictability to outsiders and reduce information risk (or estimation risk), resulting in
lower cost of equity (Francis et al. 2005). On the other hand, earnings smoothing can increase cost of equity by
reducing the quality of earnings because earnings smoothing is considered as a device for opportunistic earnings
management (Bhattacharya et al. 2003; Leuz et al. 2003; Myers et al. 2007). Motivated by these opposing views on
the relation between earnings smoothing and cost of equity, a recent study, McInnis (2008) documents no effect of
earnings smoothing on cost of equity by relying on standard asset pricing tests, suggesting that managers’ preference
on smooth earnings paths is not driven by equity market.
9
increase (decrease) the likelihood of upgrades (downgrades). Since credit rating agencies include
earnings volatility in their rating process and managers believe that smoother earnings help
achieve and preserve better credit ratings, it is likely that managers have incentives to smooth
earnings in an attempt to influence their debt ratings. However, there are several factors that
affect managers’ incentives for earnings smoothing activity.
Credit rating agencies likely have the ability to completely undo earnings management.
As argued by Jorion et al. (2005), credit rating agencies are sophisticated users of financial
information, who have access to a broad set of information, including private information. Their
access, expertise and ability weaken managers’ incentive to manage earnings. However, raters
may lack incentives to uncover bond issuers’ earnings smoothing activities because the majority
of rating agency revenue stems from rating fees paid by bond issuers. Since 1968, due to
increasing cost of maintaining the quality of credit ratings and growing demand for more rating
coverage, rating agency revenue structure has shifted from investor-based compensation (e.g.,
subscription fees collected from investors) to issuer-based payments (e.g., rating fees).
According to Moody’s 10K in 2005, it obtains 87% of its revenue from issuers. In addition,
rating agencies are developing consulting businesses which advise on matters that might affect
an issuer’s rating (Economist 2005). This may create conflicts of interest that influence agencies’
objectivity. Recent business articles also claim that the conflict of interest which exists in the
relationship between rating agencies and bond issuers is similar to that between auditors and
clients, or to the conflict of interest between stock analysts and the firms they follow.
Rating agencies are unlikely to risk their reputation for any one issuer, however.11 As
pointed out by Smith and Walter (2001) (and understood by the rating agencies), bond issuers’
11
See testimony of Sean J. Egan, Managing Director, Egan-Jones Ratings Co. before the Senate Committee on
Banking, Housing, and Urban Affairs: The Role of Credit Rating Agencies in the Capital Markets (February 8,
10
demand for rating services depends on the rating agencies’ reputation as perceived by the
investors.12 If investors view certain rating agencies as less competent and their ratings as
inaccurate, they will disregard the ratings, rendering the rating service valueless to bond issuers.
Thus, rating agencies have incentives to adjust for bond issuers’ opportunistic behaviors. This
mitigates managers’ incentives to smooth earnings, at least to the extent that rating agencies can
see through earnings smoothing activities.
Given bond issuers’ incentives to increase earnings smoothness and credit rating
agencies’ ability and incentives (or disincentives) to adjust discretionary smoothing activity,
whether bond issuers discretionarily smooth earnings remains an empirical question. Since the
implications of rating changes are greater for firms whose credit ratings are near a change to the
adjacent broad rating category, these firms would have greater incentives than others to increase
earnings smoothness. We hypothesize (in the alternative form) that:
H1: Firms whose credit ratings are near a rating upgrade or a downgrade to an adjacent
broad rating category smooth earnings smoothness to a greater extent than other firms
within the same broad rating category.
Among firms whose credit ratings are near a change to the adjacent broad rating category,
credit rating changes would result in greater benefits or costs (e.g., bond yield changes) for firms
that will rely more on debt financing in the future. These firms thus have a stronger incentive to
improve or maintain credit ratings and are more likely to smooth earnings. On the other hand,
firms that do not plan to issue a large amount of debt have weaker incentives to maintain or
improve credit ratings. This leads to the following hypothesis (in the alternative form):
2005). Also see “Report on the role and function of credit rating agencies in the operation of the securities markets,”
The Securities and Exchange Commission, January 2003.
12
This reputational concern may be even more important to rating agencies that are NRSRO’s (Nationally
Recognized Statistical Rating Organizations). These firms were designated by the SEC prior to the Credit Rating
Agency Reform Act of 2006, which supports greater competition and transparency in the credit rating industry. The
following five firms received NSRO designation: Moody’s, Standard & Poor’s, Fitch Ratings, A.M. Best, and
Dominion Bond Rating Service.
11
H2: Among firms whose credit ratings are near a rating upgrade or a downgrade to an
adjacent broad rating category, those with higher debt financing needs smooth earnings to
a larger extent.
If managers engage in earnings smoothing activity and credit rating agencies are able to
detect such behavior, agencies may adjust for the increased earnings smoothness. In this case,
smoothing behavior would not have a favorable influence on credit ratings. For example, S&P’s
methodology for non-US companies takes into account the fact that discretionary earnings
smoothing activities are more prevalent in countries outside of the U.S. “In other regions – aided
by local tax regulation - it is normal practice to take provisions against earnings in good times to
provide a cushion against downturns, resulting in a long run “smoothing” of reported profit[s].”13
However, to the extent that ratings agencies have conflicts of interest or cannot fully detect
earnings smoothing activities, ratings agencies may not completely adjust their ratings for
discretionary accrual-based increases in earnings smoothness. Thus, we investigate whether
managing earnings smoothness indeed impacts credit ratings.14
H3: Debt rating upgrades (downgrades) are more (less) likely for firms with increases in
earnings smoothness.
3. Research design
3.1 Research design for tests of H1
To test Hypothesis H1, we first examine the extent of earnings smoothing activity by
firms with a plus or minus notch rating relative to firms in the middle notch within the same
broad rating category. We consider earnings smoothness associated with discretionary accruals
as a proxy for earnings smoothing activity. We measure earnings smoothness as the standard
Corporate ratings criteria, pp. 32 from “Rating Methodology: Evaluating the Issuer” published by Standard and
Poor’s.
14
Since earnings changes influence earnings smoothness and rating agencies consider both earnings performance
and volatility in the rating process, rating agencies likely consider the interaction between these two factors. We do
not predict the impact of the interaction on the rating decision, although our research design in section 3 considers
this interaction effect.
13
12
deviation of earnings scaled by the standard deviation of cash flows from operating activities
(Hunt, Moyer, and Shevlin 1997; Leuz, Nanda, Wysocki 2003; Pincus and Rajgopal 2002;
Bowen, Rajgopal, and Venkatachalam 2008). Then this result is multiplied by -1 so that a higher
value is associated with smoother earnings. As explained by Leuz et al. (2003), scaling by cash
flow from operations controls for differences in the variability of economic performance across
firms. Such scaling also is intended to isolate any possible cash flow smoothing activity.
Earnings smoothing activity (SMOOTHACT) is measured by subtracting smoothness based on
earnings adjusted for discretionary accruals (SMOOTHXEM) from smoothness based on
reported earnings (SMOOTH). Discretionary accruals are estimated from the modified Jones
(1991) model. This measure of SMOOTHACT thus separates smoothness controlled by the
manager from inherent smoothness in earnings determined by economic and firm-specific factors.
This first measure is denoted as SMOOTHACT1, where the superscript refers to the first measure
of earnings smoothing activity.
Earnings smoothness as measured above assumes that rating agencies or managers
perceive smoothness as defined above. To mitigate the concern that this assumption could
unduly influence results, we also use the correlation between change in abnormal accruals and
change in pre-managed earnings (Tucker and Zarowin 2006) as an alternative measure of the
extent of earnings smoothing. This measure does not rely on assumptions on the form of earnings
volatility that is important to rating agencies or management. As long as management makes
discretionary accruals to reduce high or increase low pre-managed earnings, this correlation
captures smoothing activity. This second measure is denoted as SMOOTHACT2.
Another concern about our first measure of earnings smoothing activity is related to the
estimation of discretionary accruals. Kothari et al. (2005) point out that the models used in the
13
literature to estimate discretionary accruals are misspecified, especially for firms with extreme
performance. Discretionary accruals are over-estimated (under-estimated) for firms with
extremely good (bad) performance. The majority of our sample firms do not have extreme
performance, and our earnings smoothness is a multi-year measure, so the effect of this
misspecification is unlikely to be systematic and bias the results. Nonetheless, we use
performance-matched discretionary accruals in measuring earnings smoothing activity. This
measure is denoted as SMOOTHACT3.
Models (1) and (2) below test H1. Model (1) tests for differences in earnings smoothing
activity for firms with either a plus or minus notch within a broad rating category versus those in
the middle notch. Model (2) distinguishes firms with plus notches from those with minus notches.
The models are:
SMOOTHACTjt = β0 + β1D_CHGt-1 + ∑βiCONTROLi + ε
(1)
SMOOTHACTjt = β0 + β1DPLUSRt-1 + β2DMINUSRt-1 + ∑βiCONTROLi + ε
(2)
Where [COMPUSTAT item numbers in brackets]:
SMOOTHACTjt = the extent of earnings smoothing activity, proxied by the following three
measures (j=1, 2, or 3).
SMOOTHACT1t = the first measure of SMOOTHACT, calculated as the difference between
smoothness based on reported earnings (SMOOTHt) and smoothness based on earnings adjusted
for discretionary accruals (SMOOTHXEM1t). SMOOTHt and SMOOTHXEM1t are defined
below. Discretionary accruals (DA) are estimated based on the modified Jones’ (1991) model
below.
SMOOTHt = the standard deviation of earnings before extraordinary and discontinued items for
the most recent five years divided by the standard deviation of cash flow from operating
activities for the most recent five years [(-1)*s.d.(data123 for years t-4 to t) / s.d.(data308 for
years t-4 to t)].15
DAt = discretionary accruals, measured as the difference between total accruals and the
estimated non-discretionary accruals. The non-discretionary accruals model is estimated for each
15
We require at least 3 years of data to be included in the sample.
14
year and 2-digit SIC industry, based on equation (3) below. Each firm-year’s non-discretionary
accruals are then calculated based on equation (4).
SMOOTHXEM1t = earnings smoothness adjusted for discretionary smoothing activities,
measured as [(-1)*s.d.((data123-DA) for years t-4 to t)/s.d.(data308 for years t-4 to t)]. Both
earnings and cash flows are scaled by lagged total assets.16
SMOOTHACT2t = the second measure of SMOOTHACT, calculated as the correlation between
change in discretionary accruals and change in earnings adjusted for discretionary accruals (DA).
SMOOTHACT3t = the third measure of SMOOTHACT, calculated the same as the first, except
that in measuring SMOOTHXEM3t, it employs performance-matched discretionary accruals,
following Kothari et al. (2005).
SMOOTHXEM3t = earnings smoothness measured the same as SMOOTHXEM1t except that it
employs performance-matched discretionary accruals. Each sample firm is matched with a
control firm on year, industry membership based on two-digit SIC, and return on asset. The
difference in DA between each pair of sample and control firms is performance-matched
discretionary accruals.17
Total Accrualst = α0 + α1 + α2ΔSalest + α3Property Plant and Equipmentt+ ε
(3)
Non-discretionary Accrualst = ̂ 0 + ̂1 + ̂ 2 (ΔSalest - ΔAccount Receivablest) +
̂ 3 Property Plant and Equipmentt + ε
(4)
Total Accrualst is the difference between income before extraordinary items (data 123) and
operating cash flows before extraordinary item (data308-data124). All variables except the
intercept (α0) are deflated with total assets at the beginning of the year.
D_CHGt-1 =1 for firms whose prior year credit ratings are in the top or bottom category within a
broad rating, and 0 otherwise. Firms whose credit ratings are designated with a plus or a minus
notch within a broad rating are considered to be in the top or bottom category, respectively
[D_CHGt-1 = 1 if data280 in prior year equals 4, 6, 7, 9, 10, 12, 13, 15, 16, 18, 19, or 21].
DPLUSRt-1 = 1 for firms whose prior year credit ratings are in the top category within a broad rating,
i.e., credit rating designated with a plus notch, and 0 otherwise [DPLUSRt-1 = 1 if data280 in prior
year equals 4, 7, 10, 13, 16 or 19].
16
SMOOTHt for the results presented in the paper is calculated based on earnings before extraordinary and
discontinued items. Since rating agencies are likely to adjust special items reported by firms, we also calculate
SMOOTHt after adjusting for special items reported in COMPUSTAT. The results hold to this adjustment.
17
Note that due to the way SMOOTHACT2t is measured, empirically, we cannot measure earnings smoothness
adjusted for smoothing activity. Thus, there is no corresponding measure of SMOOTHXEM2t. In section 3.3 on the
tests of H3, when SMOOTHACT2t is employed, model (7) is not tested and ∆SMOOTHXEMt is omitted from
model (8).
15
DMINUSRt-1 = 1 for firms whose credit ratings are in the bottom category within a broad rating, i.e.,
credit rating designated with a minus notch, and 0 otherwise [DMINUSRt-1 = 1 if data280 equals 6, 9,
12, 15, 18, or 21].
CONTROL = control variables that proxy for earnings management capacity (NOAt-1) and
incentives unrelated to credit rating changes (BMt, DEISSt+1, LNSALESt, and ROAt), and three
sets of binary variables for broad credit rating category, annual fiscal year, and two-digit SIC code.
These variables are described in the text below.
In the empirical tests, to mitigate concerns about the undue influence of extreme
observations, we employ the ranks of SMOOTHACT1t and SMOOTHACT3t and deciles of
SMOOTHACT2t. Ranks are divided by the number of observations so that the final values fall
between zero and one.
Due to the inter-temporal nature of discretionary accruals, firms that have engaged in the
greatest extent of earnings management in the past would lack the ability to further manage
accruals. We use net operating assets (NOAt-1) to proxy for cumulative accounting management
and control for firms’ accounting discretion flexibility (Barton and Simko 2002). It is measured
at the beginning of the year as net operating assets scaled by lagged sales (data12), i.e., the sum
of shareholders’ equity (data216) and total debt (data9 + data34) minus cash and marketable
securities (data1). We also include four measures of earnings management incentives unrelated
to bond ratings. Book-to-market ratio (BMt) proxies for growth. It controls for a possible
stronger incentive of growth firms to manage earnings in order to avoid negative earnings
surprises (Skinner and Sloan 1999). The variable DEISSt+1 is a dummy variable that equals 1
when the amount of bond and equity issuance net of payment and repurchase in year t+1 is above
the median of all firm-years that report a net issue of debt or equity, and 0 otherwise. It controls
for incentives to manage earnings associated with equity or bond issuance. The logarithm of
16
sales (LNSALESt) controls for larger firms’ earnings smoothing activity induced by political
cost concerns (Watts and Zimmerman 1990).18
Because models (1) and (2) are designed to examine earnings smoothing behavior of
firms with a plus or minus notch rating relative to other firms in the same broad rating category,
we include dummy variables for broad credit rating categories to control for differences in
smoothing activity across broad categories. In addition, Jorion et al. (2008) find that bond
issuers’ earnings management activity changes over time. We include year dummies to control
for this phenomenon. We also include dummy variables for SIC code to control for variation in
earnings smoothing activity across industries. Inclusion of these two sets of binary variables
mitigates concerns about certain years/industries driving our results. Finally, we employ
clustered standard errors approach to mitigate concerns that the residuals may be correlated over
time (Petersen 2009).
A positive β1 in model (1) would provide overall support for hypothesis H1. By
examining firms with plus and minus notch ratings separately, as in model (2), we can
investigate differences in earnings smoothing activity for firms that are trying to encourage
upgrades (i.e., those with a plus notch) or discourage downgrades (i.e., those with a minus notch).
We predict a positive sign for both β1 and β2 in model (2).
3.2 Research design for tests of H2
Hypothesis H2 predicts that among companies near a rating change, those with a greater
need for future debt financing are more likely to smooth earnings. Models (5) and (6) below are
used to investigate the hypothesis.
SMOOTHACTjt = β0 + β1D_CHGHLEVt-1 + β2D_CHGLLEVt-1 + ∑βiCONTROLi + ε (5)
SMOOTHACTjt = β0 + β1D_PLUSRHLEVt-1 + β2D_PLUSRLLEVt-1 +
18
All continuous variables are winsorized at 1/99 percent. Winsorization at 2/98 percent does not change the results.
17
β3D_MINUSRHLEVt-1 + β4D_MINUSRLLEVt-1 + ∑βiCONTROLi + ε (6)
Where [COMPUSTAT item numbers in brackets]:
D_CHGHLEVt-1 = 1 for firms with D_CHGt-1 at 1 and a leverage ratio above the median
leverage for all firms with D_CHGt-1 at 1, and 0 otherwise. Leverage ratio is defined as
long-term debt divided by lagged total assets [data9 / lagged data6].
D_CHGLLEVt-1 = 1 for firms with D_CHGt-1 at 1 and a leverage ratio below or at the
median leverage for all firms with D_CHGt-1 at 1, and 0 otherwise.
D_PLUSRHLEVt-1 and D_PLUSRLLEVt-1 (D_MINUSRHLEVt-1 and D_MINUSRLLEVt-1) are
defined analogously to D_CHGHLEVt-1 and D_CHGLLEVt-1, but are for firms with a plus
(minus) notch.
We use the leverage ratio to proxy for the extent of reliance on debt financing. Leverage
is used by Bushee and Leuz (2005) to proxy for overall financing needs. In addition, Lemmon et
al. (2008) document that capital structure (i.e., leverage ratio) is very stable over time. Due to an
unobserved time-invariant effect, firms with higher (lower) leverage ratios tend to maintain
higher (lower) leverage ratios even after 20 years. Stable leverage implies that firms with higher
leverage ratios are more affected by changes in credit ratings because they continue to rely more
on debt financing in future years.
Kisgen (2006) finds that firms near rating changes manage their net debt issuance to a
greater extent than other firms. Thus, in defining high versus low leverage, we use the median
leverage of firms near rating changes instead of all sample firms. We thus categorize firms as
having high (low) financing needs if the firm’s leverage is above (below or at) the median for all
firms near rating changes. Hypothesis H2 would be supported with a finding of β1 > β2 in model
(5), and β1 > β2 and/or β3 > β4 in model (6).
3.3 Research design for tests of H3
H3 investigates the relation between changes in earnings smoothness and the probability
of future rating changes. We test hypothesis H3 with ordered probit models (7) and (8) that
18
examine the impact of change in smoothness on the likelihood of subsequent credit rating
downgrades and upgrades. Model (7) employs total change in smoothness (∆SMOOTH) while
model (8) decomposes ∆SMOOTH into ∆SMOOTHXEM and ∆SMOOTHACT to examine
whether the rating agencies distinguish the former (i.e., changes in smoothness due to changes in
underlying economic performance) from the latter (i.e., changes in smoothness due to
discretionary smoothing activities).
RTGCHGt+1 = β0 + β1∆SMOOTHt + β2D_POSt · ∆SMOOTHt + β3D_POSt +
∑βi∆CONTROLi + ε
(7)
RTGCHGt+1 = β0 + β1∆SMOOTHXEMjt + β2D_POSt · ∆SMOOTHXEMjt +
β3∆SMOOTHACTjt + β4D_POSt · ∆SMOOTHACTjt + β5D_POSt +
∑βi∆CONTROLi + ε
(8)
Where,
RTGCHGt+1 = 3 if an issuer’s credit rating is upgraded, 2 if not changed, and 1 if
downgraded in the subsequent year.
D_POSt = 1 if ROAt improves from the prior year, and 0 otherwise. ROAt is defined as
earnings before extraordinary and discontinued items divided by total assets at the
beginning of the year [data123/lagged data6].
∆SMOOTHt = rank of change in SMOOTHt, i.e., earnings smoothness based on reported
earnings as defined in model (1).
∆SMOOTHXEMjt = rank of change in SMOOTHEXMjt (j=1 or 3), i.e., earnings
smoothness based on earnings adjusted for discretionary accruals as defined in model (1).
∆SMOOTHACTjt = rank of change in SMOOTHACTjt (j=1, 2, or 3), i.e., the extent of
discretionary earnings smoothing activity as defined in model (1).
∆CONTROL = binary variables for annual fiscal years and one-digit SIC codes, annual
changes of leverage (∆LEVt), revenue (∆SALESt), free cash flows (∆FCFt), capital
intensity (∆PPEt), firm size (∆SIZEt), and interest coverage (∆INT_COVt).19
LEV = long-term debt divided by lagged total assets [data9 / lagged data6].
SALES = sales divided by lagged total assets [data12 / lagged data6].
19
Since not all industries classified based on the two-digit SIC codes experience rating changes in our sample period,
we use the one-digit SIC code instead.
19
FCF = cash flow from operating activities minus average capital expenditure over current
and past two years divided by lagged total assets [(data308-average of data128) / lagged
data6].
PPE = gross PPE divided by lagged total assets [data7 / lagged data6].
SIZE = natural log of total assets [data6].
INT_COV = operating income before depreciation divided by interest expense [data13 /
data15].
Similar to previous models, models (7) and (8) use the rank of ∆SMOOTHt and ranks of
∆SMOOTHXEMjt and ∆SMOOTHACTjt, respectively, rather than the continuous measures to
mitigate undue influence of extreme values and to reflect a possible nonlinear relation. In the
models, D_POSt controls the impact of performance changes in the rating revision process.
While performance changes likely have an impact on income smoothing, higher volatility caused
by earnings improvements presumably will not decrease (increase) the likelihood of an upgrade
(downgrade). We therefore interact D_POSt with ∆SMOOTHt in model (7) and both
∆SMOOTHXEMjt and ∆SMOOTHACTjt in model (8) and expect the interaction terms to have
signs opposite from the coefficients on ∆SMOOTHt, ∆SMOOTHXEMjt, and ∆SMOOTHACTjt.
Based on prior research (e.g., Ashbaugh-Skaife et al. 2006) and the rating methodologies of S&P
and Moody’s, both models also include changes in leverage, sales, free cash flows, capital
intensity, firm size, and interest coverage as controls. Decreases in free cash flow, sales, and
interest coverage, and increase in leverage proxy for higher default risk. Larger firm size and
higher capital intensity represent lower risk.
4. Sample, descriptive statistics, and empirical results
4.1. Sample and descriptive statistics
20
The sample includes all bond issuers available in COMPUSTAT from 1990 to 2005 that
are not in the utility or financial industries.20 Sample firms are required to have a credit rating by
S&P ranging from AA+ to CCC- and necessary COMPUSTAT data used in calculating measures
of earnings smoothing activities and control variables. We eliminate bond issuers in industries
with fewer than ten annual observations to estimate the modified Jones (1991) model. Variables
used in estimating the non-discretionary level of accruals are winsorized at three standard
deviations by each year (e.g., Tucker and Zarowin 2006). For our main tests, the final sample
ranges from 8,542 (when SMOOTHACT2 is used) to 9,437 (when SMOOTHACT3 is used) firmyear observations.
Panel A of table 1 lists the sample distribution by fiscal year and panel B presents the
sample distribution and the mean/median extent of smoothing (SMOOTHACTj) by credit rating.
Panel A shows that the number of observations is gradually increasing over time from 1990 to
2005. Only ratings B+ and BBB represent 10 percent or more of the sample, with 11.8% and
10.0%, respectively. Overall, the sample distribution across the rating categories shows that any
results are unlikely to be driven by a single rating category. Panel B also displays the relative
extent of earnings smoothing activities (SMOOTHACTj) based on all three measures by credit
ratings. Within the same broad rating category, in general, firms with a plus or minus notch tend
to smooth earnings to a larger extent than those in the middle notch. This tendency is more
pronounced within investment grade firms.
Panel C of table 1 presents the mean, median, and standard deviation of each variable
used in the tests for the three groups: the total sample, firms that have a middle notch rating, and
firms that have a plus or minus notch rating. The sample has a wide distribution in all variables
20
As 1988 is the first year when all firms have statement of cash flow data, and at least three consecutive years are
required in measuring earnings smoothing activities, our sample period starts in 1990.
21
except binary variables of UPGRADED (with value 1 for firms experiencing an upgrade) and
DOWNGRADED (with value 1 for firms experiencing a downgrade), as reflected in the large
standard deviations. Between the two subgroups, firms in the middle notch are involved in
significantly less earnings smoothing activity relative to firms with a plus or minus notch rating.
The middle-notch firms also have lower book-to-market ratios, larger sales, lower net operating
assets, and lower likelihood of being upgraded compared to the outer-notch firms.
Figure 1 depicts the relative extent of earnings smoothing activities based on both the
mean and median of SMOOTHACT1 (which uses discretionary accruals estimated from the
modified Jones (1991) model) within each broad category, respectively. Out of the six broad
rating categories, four in both panels show the predicted pattern, in which observations near a
broad rating change smooth their earnings to a larger extent than do other firms in the same
rating category. This, together with table 1 provides preliminary support for H1. In the figures,
the solid line divides the categories into investment grade (to the left) and speculative grade (to
the right). The patterns for speculative-grade observations are inconsistent with H1 for the B
rating category and the minus notch of CCC rating category. It thus appears that smoothing
behavior may differ for investment- versus speculative-grade categories.21
To show that the pattern of SMOOTHACTt we observe in figure 1 is not driven by cash
flow smoothing, in figure 2, we separately display the volatility of earnings and cash flows in
panels A and B, respectively. Figure 2 does not exhibit the same pattern as figure 1. This
demonstrates that the greater extent of earnings smoothing in firms with a minus or a plus notch
that is depicted in figure 1 is not caused by cash flow volatility. Overall, figures 1 and 2 show
21
In Section 5, we further investigate whether differences in earnings smoothing behavior exist between investment
and speculative grade categories based on multivariate analysis and provide potential explanations for why earnings
smoothing behavior may be different between two groups.
22
that firms near broad rating changes smooth earnings to a larger extent than other firms and this
pattern is not driven by cash flow volatility.
4.2. Extent of earnings smoothing
Multivariate regressions based on models (1) and (2) control for variation in discretionary
accruals by rating category, year, industry, earnings management capacity, and incentives
unrelated with maintaining or improving credit ratings. Panels A, B, and C of table 2 present
results for all three measures of SMOOTHACTj. SMOOTHACT1 for panel A is based on
discretionary accruals estimated from the modified Jones (1991) model. The coefficient on the
dummy variable for firms with a plus or minus notch rating, D_CHGt-1, is significantly positive,
consistent with firms in this group engaging in significantly greater extent of earnings smoothing.
Further, this difference in earnings smoothing activities is present in both firms with a plus notch
rating and those with a minus notch rating relative to middle notch firms, as reflected in the
positive coefficients on DPLUSRt-1 and DMINUSRt-1 in model (2).
Results in panel B are based on SMOOTHACT2, measured as decile of the correlation
between change in discretionary accruals and change in pre-managed earnings. We employ
ordered probit approach to estimate models (1) and (2). Results are consistent with those in panel
A, albeit somewhat weaker for the coefficient on DPLUSRt-1. Thus, our results are robust to the
form of earnings smoothness as defined above. Panel C of table 2 presents results of estimating
models (1) and (2) with SMOOTHACT3, measured based on performance-matched discretionary
accruals. The results are consistent with those in panel A except that the coefficient on DMINUSRt-1
becomes insignificant.22
22
Results of testing H2, presented in Table 3, suggest that the lack of statistical significance in the coefficient on
DMINUSR is driven by firms with lower reliance on debt financing.
23
Overall, the results are consistent with the general patterns described in table 1 and figure
1 and support H1 that firms with ratings near a broad rating change smooth earnings to a
significantly larger degree than other firms in the same broad rating category.
4.3. Extent of earnings smoothing by debt financing needs
Table 3 presents results for models (5) and (6). To test H2, we compare the coefficients
on D_CHGt-1, D_PLUSRt-1, and D_MINUSRt-1 for firms with high versus low leverage. Panel B of
table 3 reports p-values of F-tests on differences in the coefficients between the two groups.
Results of model (5) using both the second and third measures of SMOOTHACTjt show that
among firms near rating changes, those with higher leverage smooth earnings to a greater extent,
as reflected in both the magnitude and statistical significance of the coefficients on D_CHGHLEVt-1
versus D_CHGLLEVt-1..In particular, the coefficients on D_CHGHLEVt-1 (firms with higher leverage
and near broad rating change) are significant at the 1% level while the coefficients on
D_CHGLLEVt-1 (firms with lower leverage and near broad rating change) are insignificant. When
SMOOTHACT2 is used, the difference in the coefficients for D_CHGHLEVt-1 versus D_CHGLLEVt-1
is 0.060 with a p-value of 0.023. Estimation results based on SMOOTHACT3 are stronger; the
difference in the coefficients on D_CHGHLEVt-1 versus D_CHGLLEVt-1 is 0.016 with a p-value of
0.016. However, when SMOOTHACT1 is used, the difference is insignificant (p-value = 0.19).
Results for model (6), which separates firms with plus versus minus notches, show that
the difference is primarily driven by firms with minus notches. When SMOOTHACT1 is used for
model (6), the coefficient on D_MINUSRHLEVt-1 is greater than that on D_MINUSRLLEVt-1 by 0.016
with a p-value of 0.051 while the coefficient on D_PLUSRHLEVt-1 is not significantly different
from that on D_PLUSRLLEVt-1. Estimation results based on SMOOTHACT3 are more pronounced;
the difference in the coefficients on D_MINUSRHLEVt-1 versus D_MINUSRLLEVt-1 is 0.026 with a p-
24
value of 0.006 while the coefficient on D_PLUSRHLEVt-1 is again not significantly different from
that on D_PLUSRLLEVt-1. However, when SMOOTHACT2 is used, coefficients on both
D_PLUSRHLEVt-1 and D_MINUSRHLEVt-1 are greater than those on D_PLUSRLLEVt-1 and
D_MINUSRLLEVt-1 by 0.058 and 0.068 with p-values of 0.079. Overall, the evidence in table 3
supports H2, that among firms near rating changes, those with higher debt financing needs
engage in earnings smoothing activity to a larger extent than those with lower debt financing
needs. The results are more pronounced for firms that are more likely to be downgraded to a
lower broad rating category and thus face higher future borrowing costs.
4.4. The impact of smoothness changes on the likelihood of credit rating changes
Hypothesis H3 addresses the impact of smoothness changes on future credit rating
changes. Regression results for model (7) are reported in panels A and C of table 4, where
changes in smoothness can be calculated. Again, unlike SMOOTHACT1 and SMOOTHACT3 ,
SMOOTHACT2 does not have corresponding variables, ∆SMOOTHt and ∆SMOOTHXEMt.
Results in table 4 indicate that, controlling for increases in ROAt (D_POSr = 1), improvement in
earnings smoothness decreases (increases) the likelihood of rating downgrades (upgrades).23 The
results for model (8), which separates smoothing into discretionary and non-discretionary
components, indicate that both components of earnings smoothness decrease (increase) the
likelihood of rating downgrades (upgrades). The interaction terms for change in ROAt and
change in smoothness are consistently opposite in sign to smoothness changes and significantly
different from zero. Unreported chi-square tests show that the sums of the smoothing measures
23
In the estimation of models (7) and (8), we only include rating changes of one notch. This mitigates concern that
multiple-notch changes are likely caused by significant changes in the firm’s operating environment or other factors
that do not relate to earnings. In addition, using only one-notch changes demonstrates how firms’ smoothing activity
affects the probability of being upgraded or downgraded to the adjacent broad rating category. We also use logistic
regressions that separately investigate the likelihood of upgrades and downgrades. Confirming results in table 4,
improvement in earnings smoothness decreases the likelihood of rating downgrades and increases the likelihood of
rating upgrades.
25
and their associated interaction terms are generally not significantly different from zero. This
suggests that the impact of smoothing on future rating changes is not driven by the changes in
smoothness associated with increases in ROAt. Results for SMOOTHACT2, reported in Panel B,
are similar. Overall, the results in table 4 support H3 and indicate a significantly positive impact
on future credit ratings of earnings smoothness for companies experiencing decreases in ROA.24
5. Supplemental tests
5.1. The absolute value of discretionary accruals as a measure of accounting discretion
The literature employs the absolute magnitude of discretionary accruals (|DA|) to
measure the extent of accounting discretion managers exert to report certain levels of earnings
(e.g., Warfield, Wild, and Wild 1995; Frankel, Johnson, and Nelson 2002; Klein 2002; Bowen,
Rajgopal, and Venkatchalam 2008). Following the literature and to gauge the magnitude of
accounting discretion managers exert to arrive at desired earnings smoothness, we replace the
measure of earnings smoothing activity with |DA| in our models as the dependent variable. The
advantage of using |DA| is that it does not require any assumptions about the functional form of
earnings smoothness used by management. As long as management engages in accruals
management to improve earnings smoothness, the absolute amount of discretionary accruals will
reflect earnings smoothing activity. The drawback of |DA| is that it is a one-year measure and
does not fully correspond to the longer-term behavior inherent in earnings smoothing.
Untabulated results for models (1) and (2) show that |DA| for firms with either a plus or a minus
notch rating is larger than |DA| for firms in the middle notch, suggesting that managers of the
former exert accounting discretion to a larger extent. Results for models (5) and (6) show that
among firms with a plus or minus notch rating, |DA| does not vary with leverage. Although the
24
The differential impact of change in earnings smoothness when earnings increase versus when earnings decrease
suggests stronger incentives for management to smooth earnings in the latter scenario. In Section 5.2, we further
examine whether the extent of earnings smoothing differs in these two situations.
26
coefficient on D_PLUSRHLEV (firms with a plus notch rating and higher leverage) is significantly
positive, its magnitude is not significantly different from that on D_PLUSRLLEV (firms with a
plus notch rating and lower leverage).25
5.2. Earnings smoothing activity by earnings thresholds
This section examines earnings smoothing activity separately for firms that meet or beat
earnings thresholds versus those that miss the same thresholds. This analysis would provide
evidence on whether empirical evidence in Section 4, i.e., a larger extent of earnings smoothing
for firms with greater incentives to maintain or improve their credit ratings, is conditional on
surpassing earnings thresholds.
In the setting of an equity market, Kirschenheiter and Melumad (2002) analytically show
that firms take a “big bath” when earnings are inferior and smooth earnings when earnings are
superior. Their findings suggest that earnings smoothing activities may be more prevalent for
firms that surpass earnings thresholds than for firms missing earnings thresholds. However,
management’s tendency to engage in the two types of activities likely differs in debt markets.
Unlike stockholders, creditors have a fixed claim to a firm’s value and do not share the upside
growth potential (Fischer and Verrecchia 1997; Plummer and Tse 1999). Thus, creditors are
more concerned about performance deterioration than growth (Jiang 2008), implying that the
incentives to smooth earnings still exist even when earnings are inferior – e.g., when earnings
miss a threshold.
Table 5 presents the results of estimating models (1) and (2) conditional on
meeting/beating or missing three earnings thresholds – earnings increases, positive earnings, and
25
Hribar and Nichols (2007) document that the absolute amount of discretionary accruals is correlated with the
volatility of cash flows. Following them, we include the volatility of assets-scaled cash flows from operating activity
in models when |DA|t is the dependent variables. Results are robust to the inclusion of this additional control
variable.
27
meeting analysts’ expectation.26 For these tests, we report results only for our variables of
interest. Panel A of table 5 shows results for models (1) and (2) for earnings increase and
decrease subsamples. With the exception of firms with a minus notch rating that report an
earnings increase, for both subsamples, firms near rating changes smooth earnings to a larger
extent than others in the same broad rating category. This suggests that firms’ earnings
smoothing decisions are not conditional on earnings changes. Results in Panel B for model (1)
indicate that in both loss and profit firms, those near rating changes smooth earnings to a larger
extent than others in the same broad rating category. Results for model (2) for these subsamples
are mixed, however. Among loss firms, only those with minus notch ratings engage in more
earnings smoothing activities than firms in the middle notch, while in the subsample of firms
reporting positive earnings, only firms with plus notch ratings smooth earnings to a larger extent
than other firms. Results reported in Panel C for meeting or beating versus missing analysts’
forecasts show that among firms missing analysts’ forecasts, only firms with plus notch ratings
engage in more earnings smoothing activities than firms in the middle notch. For the subsample
of firms meeting or beating analysts’ forecasts, firms with both plus and minus notch ratings
engage in more earnings smoothing activities. In sum, our hypothesis that firms close to a rating
change smooth earnings to a larger extent is generally supported irrespective of earnings
thresholds, but the results are somewhat weaker for firms missing analysts’ forecasts.
We also investigate whether firms near rating changes are more likely to: (1) meet or beat
the three earnings thresholds; and (2) engage in more income-increasing earnings management
(i.e., higher magnitude of discretionary accruals) to achieve earnings targets. We find (in
untabulated results) that firms near rating changes are not more likely to meet earnings
26
Results in table 5 are based on SMOOTHACT1 as the dependent variable. Main implications are similar with the
other two measures of earnings smoothing activities.
28
thresholds nor to report a higher magnitude of signed discretionary accrual. A possible
interpretation is that since meeting or beating earnings thresholds is more important in equity
markets than in debt markets, the extent of income-increasing discretionary accruals to achieve
earnings targets are similar to both firms near rating changes and the other firms in the same
broad rating category. On the other hand, since the stock market rewards for smoother earnings
path (e.g., lower cost of equity capital) are trivial (McInnis 2008), the difference in earnings
smoothing activities between firms near rating changes and the other firms is more pronounced
in the setting of debt markets.
5.3. Difference in earnings smoothing between investment vs. speculative grade firms
The descriptive statistics in table 1 and figure 1 suggest a different pattern between firms
with investment-grade ratings and those with speculative-grade ratings. To incorporate this
possible difference, we augment models (1) and (2) by including a dummy variable for
speculative-grade and interacting the dummy variables with D_CHG, DPLUSR, and DMINUSR.
Untabulated results of the augmented models (1) and (2) show a larger extent of earnings
smoothing activities for firms near rating changes within investment-grade ratings than for those
within speculative-grade. In addition, when using only the investment-grade rating firms, the
coefficients on D_CHG, DPLUSR, and DMINUSR become larger.
We explore three plausible explanations for the weaker results among firms with noninvestment grade ratings. First, the difference could be due to a higher likelihood of having a
split rating for the non-investment grade firms. A split rating occurs when the rating agencies
(e.g., S&P and Moody’s) assign different ratings to the same security. For example, for a firm
which is assigned a plus (or minus) notch rating by both S&P and Moody’s, the incentive to
improve (or maintain) credit ratings is presumably greater than that for a firm receiving a plus (or
29
minus) notch rating by one rating agency, but a middle notch by the other agency. If noninvestment grade firms are more likely to have split credit ratings by S&P and Moody’s due to
more uncertain information environment of bond issuers than investment grade firms, weaker
results for non-investment grade firms corroborate our findings in section 4. To investigate this
possibility, we estimate how often Moody’s and S&P assign the same notch ratings. We find that
based on FISD database, for investment grade ratings, more than 79% of bond issues are
assigned the same notch ratings. On the other hand, for non-investment grade ratings, only 54%
of bond issues have the same notch ratings by both rating agencies.27
Second, weaker results for speculative grade firms could also be attributed to the fact that
multiple-notch rating revisions are more likely within speculative grade ratings. Based on the
FISD database, we find that 73 percent (55 percent) of bond issues within investment (noninvestment) grade ratings are revised by one notch, suggesting that for firms with a middle notch
rating, those in the non-investment grade are more likely to be downgraded or upgraded to the
next broad rating category than those in the investment grade. This implies that the incentive to
maintain or improve credit ratings is less distinct between firms near rating change and firms in
the middle notch within the non-investment grade. This finding also corroborates our results in
section 4. Third, accounting flexibility to manage earnings is higher for investment-grade firms
due to the scale of their operations, which makes it easier to shift income across operations and
time (Jorion et al. 2008). Therefore, although the incentive to smooth earnings to manage credit
ratings also exists for the non-investment grade firms, the lack of accounting flexibility may
constrain firms near rating changes from engaging in significantly larger extent of earnings
smoothing than other firms.
27
This result is based on all bond issues of 217,483 from FISD database. More detailed descriptive statistics can be
provided upon request. We also provide details about the FISD database in the appendix.
30
6. Conclusion
The importance of credit ratings perceived by both the bond and stock market and, thus,
by bond issuers, suggests the existence of incentives for bond issuers to improve or maintain
their credit ratings. We investigate earnings smoothing as one way in which bond issuers affect
credit ratings and argue that bond issuers with greater incentives to improve or maintain credit
ratings will smooth earnings to a larger extent than other issuers. Further, we examine whether
these incentives are stronger for firms that rely more on the debt market for financing. Finally,
we analyze whether bond issuers’ earnings smoothing activity is effective at changing the
likelihood of subsequent rating upgrades or downgrades.
Drawing on the literature and empirical evidence provided in the appendix to this study,
we categorize bond issuers whose credit ratings have a plus or minus notch as having greater
incentives to improve or maintain bond ratings. We find that earnings smoothing via earnings
management is more concentrated in these firms. Among the group of firms with a plus or minus
notch rating, those with higher debt financing needs smooth earnings to a larger extent than
others. Our results are robust to including other factors that influence earnings management and
additional sensitivity tests. In addition, we find that controlling for earnings smoothness related
to an increase in ROAt, earnings smoothing activity indeed increases (decreases) the likelihood
of a subsequent rating upgrade (downgrade). Thus, it appears that earnings smoothing activity
can be used as a tool to manage credit ratings.
This paper provides fresh empirical evidence on the earnings smoothing incentives of firms
that participate in the public debt market. It is among the first to examine a setting where we can
identify companies that have stronger incentives to manage earnings to affect debt ratings. We
empirically document that rather than being ubiquitous, earnings smoothing activities vary
31
significantly with these incentives. Our results also suggest that not only do managers smooth
earnings to influence credit ratings, but also that credit rating agencies do not fully compensate
for such smoothing in their credit rating decisions.
32
References
Ashbaugh-Skaife, H., D. Collins, and R. LaFond (2006): “The Effects of Corporate Governance on
Firms’ Credit Ratings,” Journal of Accounting & Economics 42 (1/2): 203-243.
Barnea, A., J. Ronen and S. Sadan (1975): “The Implementation of Accounting Objectives: An
Application to Extraordinary Items,” The Accounting Review.
Barton, J., and P. J. Simko (2002): “The Balance Sheet as an Earnings Management Constraint,” The
Accounting Review 77 (4): 1-27.
Beaver, W., P. Kettler, and M. Scholes (1970): “The Association Between Market Determined and
Accounting Determined Risk Measures,” The Accounting Review 45(4): 654-682.
Bhattacharya, U., H. Daouk, and M. Welker (2003): “The World Price of Earnings Opacity,” The
Accounting Review 78: 641-678.
Blume, M.E., F. Lim, and A. C. MacKinlay (1998): “The Declining Credit Quality of US Corporate Debt:
Myth or Reality?” Journal of Finance 53: 1389-1413.
Bushee, B., and C. Leuz (2005): "Economic Consequences of SEC Disclosure Regulation: Evidence from
the OTC Bulletin Board," Journal of Accounting and Economics 39: 233-264
Bowen, R., S. Rajgopal, and M. Venkatachalam (2008): “Accounting Discretion, Corporate Governance,
and Firm Performance.” Contemporary Accounting Research, 25 (2): 310-405.
Clarkson, B. (2007): “Transparancy and Trust Must Keep on Driving Rating Agencies – Brian ClarksonINSIGHT,” Financial Times London Ed1, 9/18, p42.
Collins, D., D. Dhaliwal, and M. Rozeff (1981): "The Economic Determinants of the Market Reaction to
Proposed Mandatory Accounting Changes in the Oil and Gas Industry: A Cross-Sectional
Analysis," Journal of Accounting and Economics 3: 37-71.
Cook, R.D. (1977): "Detection of Influential Observations in Linear Regression," Technometrics, 19: 15 18.
Covitz, D.M. and P. Harrison (2003): “Testing Conflicts of Interest at Bond Ratings Agencies with
Market Anticipation: Evidence that Reputation Incentives Dominates,” working paper, Federal
Reserve Board.
Dechow, P., R. Sloan, and A. Sweeney (1995): “Detecting earnings management,” The Accounting
Review 70: 193-225.
Dechow, P. M., S. A. Richardson, and I. Tuna (2003): “Why Are Earnings Kinky? An Examination of the
Earnings Management Explanation,” Review of Accounting Studies 8: 355-384.
Dichev, Ilia D., and J. D. Piotroski (2001): “The Long-Run Stock Returns Following Bond Ratings
Changes,” Journal of Finance 56 (February): 173-204.
33
Economist (2005): “Special report: Who rates the raters?” 3/26/2005, Vol. 374 Issue 8419, p67-69
Ederington, Louis H. and Jeremy C. Goh (1998): “Bond Rating Agencies and Stock Analysts: Who
Knows What When?” Journal of Financial and Quantitative Analysis 33 (December): 569-585.
Fama, E.F. and J.D. MacBeth (1973): “Risk, Return, and Equilibrium: Empirical Tests,” Journal of
Political Economy 81.
Fisher, P., and R. Verrecchia (1997): “The effect of limited liability on the market response to
disclosure,” Contemporary Accounting Research 14: 515–541.
Francis, J., R. LaFond, P. Olsson, and K. Schipper (2005): “The Market Pricing of Accruals Quality,”
Journal of Accounting and Economics 39 (2): 295-327.
Frankel, R., M. Johnson, and K. Nelson (2002): “The Relation between Auditors’ Fees for Nonaudit
Services and Earnings Management,” The Accounting Review 77: 71–105.
Fudenberg, D. and J. Tirole (1995): "A Theory of Income and Dividend Smoothing," Journal of Political
Economy, 103: 75-93.
Goel, A.M. and A.V. Thakor (2003): “Why Do Firms Smooth Earnings?” Journal of Business 76: 151192.
Goh, J.C. and L.H. Ederington (1993): “Is a Bond Rating Downgrade Bad News, Good News, or no
News for Stockholders?” Journal of Finance 48: 2001-2008.
Goh, J.C. and L.H. Ederington (1999): “Cross-Sectional Variation in the Stock Market Reaction to Bond
Rating Changes,” Quarterly Review of Economics and Finance 39: 101-112.
Gonedes, N.J. (1972): “Income-smoothing Behavior under Selected Stochastic Processes,” Journal of
Business 45.
Graham, John R., Cam Harvey and Shiva Rajgopal (2005): “The Economic Implications of Corporate
Financial Reporting,” Journal of Accounting and Economics 40: 3-73.
Gu, Z. and Y. Zhao (2006): “Accruals, Income Smoothing and Bond Ratings,” Working paper.
Holthausen, R.W. and R.W. Leftwich (1986): “The Effect of Bond Rating Changes on Common Stock
Prices,” Journal of Financial Economics 17: 57-89.
Hribar, P., and D. C. Nichols (2007): “The Use of Unsigned Earnings Quality Measures in Tests of
Earnings Management,” Journal of Accounting Research 45: 1017 – 1053.
Hunt, A., S. Moyer, and T. Shevlin (1997): “Eamings volatility, eamings management, and equity value,”
Working paper. University of Washington.
Jiang, J. (2008): “Beating Earnings Benchmarks and the Cost of Debt,” Accounting Review 83 (2): 377416.
34
Jones, J. (1991): “Earnings management during import relief investigations,” Journal of Accounting
Research 29 (Autumn): 193-228.
Jorion, P., Z. Liu, and C. Shi (2005): “Informational Effects of Regulation FD: Evidence from Rating
Agencies,” Journal of Financial Economics 76: 309-330.
Jorion, P., C. Shi, and S. Zhang (2008): “Tightening Credit Standards: The Role of Accounting Quality,”
Review of Accounting Studies, forthcoming.
Jorion, P. and G. Zhang (2007): “Non-Linear Effects of Bond Ratings Changes: The Role of the Rating
Prior to the Announcement,” Journal of Fixed Income 16 (4): 45-59.
Kisgen, D. (2006): “Credit Ratings and Capital Structure,” Journal of Finance Vol LXI(3): 1035-1072.
Klein, A. (2002): “Audit Committee, Board of Director Characteristics and Earnings Management,”
Journal of Accounting and Economics 33: 375–400.
Kothari, S. P., A. J. Leone and C. E Wasley (2005): “Performance Matched Discretionary Accrual
Measures,” Journal of Accounting and Economics 39 (1): 163-197.
Lemmon, M.L., M. R. Roberts, and J. F. Zender (2008): “Back to the Beginning: Persistence and the
Cross-Section of Corporate Capital Structure,” Journal of Finance, forthcoming.
Leuz, C., D. Nanda, and P. D. Wysocki (2003): “Earnings Management and Investor Protection: An
International Comparison,” Journal of Financial Economics 69: 505–527.
Lys, T. (1984): “Mandated Accounting Changes and Debt Covenants: the Case of Oil and Gas
Accounting,” Journal of Accounting and Economics 6: 39 – 65.
McInnis, J. (2008): “Are Smoother Earnings Associated With a Lower Cost of Equity Capital?” Working
paper.
Myers, J., L.Myers, and D. Skinner (2007): “Earnings Momentum and Earnings Management,” Journal
of Accounting, Auditing, & Finance 22: 249-284.
Petersen, M. (2009): “Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches,”
Review and Financial Studies, Volume 22: 435-480.
Pincus, M., and S. Rajgopal (2002): “The interaction between accrual management and hedging:
Evidence from oil and gas firms,” The Accounting Review 11 (1): 121-60.
Plummer, C., and S. Tse (1999): “The effect of limited liability on the informativeness of earnings:
evidence from the stock and bond markets,” Contemporary Accounting Research 16 (3): 541–574.
Rogers, William (1993): “Regression Standard Errors in Clustered Samples,” Stata Technical Bulletin 13,
pp. 19-23.
Shah, R. (2007): “Do Firms Reduce Investment to Avoid Credit Rating Downgrades?” working paper,
University of Texas at Austin.
35
Smith, R.C. and I. Walter (2001): “Rating Agencies: Is There an Agency Issue,” working paper, New
York University.
Trueman, B. and S. Titman (1988): “An Explanation for Accounting Income Smoothing,” Journal of
Accounting Research 26 (supplement): 127-139.
Tucker, J. W. and P. A. Zarowin (2006): Does Income Smoothing Improve Earnings Informativeness?
The Accounting Review 81: 251–270.
Watts, R.L. and J. L. Zimmerman (1986): “Positive Accounting Theory,” Englewood Cliffs, NJ: Prentice
Hall.
Warfield, T., J. Wild, and K. Wild (1995): “Managerial Ownership, Accounting Choices, and
Informativeness of Earnings,” Journal of Accounting and Economics 20: 61–91.
36
Table 1: Distribution of Sample by Fiscal Years and Credit Ratings
Panel A: Distribution by fiscal year
Year
N
Percent
1990
115
1.110
1991
490
4.710
1992
542
5.210
1993
533
5.120
1994
519
4.990
1995
538
5.170
1996
592
5.690
1997
610
5.860
Year
N
Percent
1998
646
6.210
1999
657
6.320
2000
716
6.880
2001
828
7.960
2002
903
8.680
2003
934
8.980
2004
927
8.910
2005
852
8.190
Mean
SMOOT
HACT2t
0.495
0.391
0.472
0.534
0.501
0.515
0.536
0.497
0.536
0.512
0.476
0.513
0.468
0.465
0.465
0.565
0.516
0.467
Median
SMOOT
HACT2t
0.444
0.333
0.444
0.556
0.444
0.444
0.556
0.444
0.556
0.556
0.444
0.556
0.444
0.444
0.444
0.556
0.556
0.389
Mean
SMOOT
HACT3t
0.566
0.464
0.531
0.554
0.500
0.515
0.517
0.487
0.524
0.491
0.502
0.501
0.480
0.464
0.450
0.470
0.453
0.441
Median
SMOOT
HACT3t
0.559
0.458
0.532
0.568
0.488
0.514
0.518
0.483
0.544
0.485
0.498
0.518
0.478
0.439
0.430
0.511
0.481
0.447
Panel B: Earnings smoothing activities by credit ratings
Rating
AA+
AA
AAA+
A
ABBB+
BBB
BBBBB+
BB
BBB+
B
BCCC+
CCC
CCC-
N for
SMOOT
HACT1t
66
274
313
571
935
737
861
1041
888
595
841
1004
1224
612
234
108
67
31
Percent
for
SMOOT
HACT1t
0.6%
2.6%
3.0%
5.5%
9.0%
7.1%
8.3%
10.0%
8.5%
5.7%
8.1%
9.7%
11.8%
5.9%
2.2%
1.0%
0.6%
0.3%
Mean
SMOOT
HACT1t
0.559
0.429
0.485
0.528
0.499
0.523
0.534
0.505
0.534
0.510
0.478
0.518
0.466
0.474
0.444
0.474
0.434
0.381
Median
SMOOTH
ACT1t
0.575
0.370
0.462
0.521
0.472
0.514
0.555
0.496
0.556
0.521
0.466
0.545
0.461
0.482
0.455
0.528
0.432
0.355
37
Panel C: Descriptive statistics of variables
Firms in the middle notch
(D_CHG=0)
Total sample
Variable
N
Mean
Median
S.D.
N
SMOOTHACT1t
10402
0.500
0.500
0.289
3770
0.486***
SMOOTHACT2t
8634
0.500
0.500
0.319
3191
SMOOTHACT
3
Mean Median
Firms with a plus/minus notch
(D_CHG=1)
S.D.
N
0.473
0.289
6632
0.508
0.516
0.288
0.481***
0.444
0.321
5443
0.511
0.556
0.318
***
S.D.
0.481
0.289
6697
0.507
0.512
0.289
10491
0.500
0.500
0.289
3794
BMt
9400
0.380
0.434
1.410
3469
0.334**
0.421
1.570
5931
0.407
0.439
1.307
LNSALESt
10392
7.688
7.630
1.442
3766
7.830***
7.800
1.462
6626
7.607
7.538
1.425
DEISSt+1
10402
0.169
0.000
0.375
3770
0.175
0.000
0.380
6632
0.166
0.000
0.372
***
0.617
0.736
6623
0.882
0.641
0.816
t
0.488
Mean Median
NOAt-1
10383
0.863
0.633
0.788
3760
0.828
UPGRADED
9199
0.075
0.000
0.263
3357
0.063***
0.000
0.243
5842
0.082
0.000
0.274
DOWNGRADED
9199
0.123
0.000
0.328
3357
0.117
0.000
0.321
5842
0.126
0.000
0.332
ΔLEVt
9199
-0.008
-0.009
0.223
3357
-0.006
-0.007
0.246
5842
-0.010
-0.009
0.208
ΔSALESt
9199
-0.005
0.009
0.357
3357
-0.005
0.007
0.381
5842
-0.005
0.009
0.343
ΔFCFt
8997
0.003
0.002
0.076
3272
0.002
0.002
0.085
5725
0.003
0.002
0.071
D_POSt
9199
0.532
1.000
0.499
3357
0.533
1.000
0.499
5842
0.531
1.000
0.499
ΔPPEt
9199
-0.005
0.006
0.231
3357
-0.004
0.004
0.232
5842
-0.005
0.007
0.230
ΔSIZEt
9199
0.060
0.041
0.216
3357
0.060
0.044
0.211
5842
0.061
0.040
0.219
ΔINT_COVt
9054
1.468
0.263
74.522
3305
0.409
0.273
34.241
5749
2.078
0.260
89.843
ΔROAt
9199
0.000
0.002
0.086
3357
0.001
0.002
0.083
5842
0.000
0.002
0.088
38
Note:
***, **, and * denote that the difference of the mean between firms with a plus or minus notch rating and those with a middle notch rating is significant at 0.01,
0.05, and 0.10, respectively.
D_CHGt-1=1 if data280 equals to 4, 6, 7, 9, 10, 12, 13, 15, 16, 18, 19, or 21, and 0 otherwise.
SMOOTHACT1t = the first measure of SMOOTHACT, calculated as the difference between smoothness based on reported earnings (SMOOTH t) and
smoothness based on earnings adjusted for discretionary accruals (SMOOTHXEM 1t). SMOOTHt = the standard deviation of earnings before
extraordinary and discontinued items for the most recent five years divided by the standard deviation of cash flow from operating activities for the most
recent five years [(-1)*s.d.(data123 for years t-4 to t) / s.d.(data308 for years t-4 to t)]. SMOOTHXEM1t = earnings smoothness adjusted for
discretionary smoothing activities, measured as [(-1)*s.d.((data123-DA) for years t-4 to t) / s.d.(data308 for years t-4 to t)]. Both earnings and cash
flows are scaled by lagged total assets.
DAt = discretionary accruals, measured as the difference between total accruals and the estimated non-discretionary accruals. The non-discretionary accruals
model is estimated for each year and 2-digit SIC industry, based on the equation below.
Total Accrualst = α0 + α1 + α2ΔSalest + α3Property Plant and Equipmentt+ ε
Total Accrualst is the difference between income before extraordinary items (data 123) and operating cash flows before extraordinary item
(data308-data124). All variables except the intercept (α0) are deflated with total assets at the beginning of the year.
Each firm-year’s Non-discretionary Accrualst = ̂ 0 + ̂1 + ̂ 2 (ΔSalest - ΔAccount Receivablest) + ̂ 3 Property Plant and Equipmentt + ε
SMOOTHACT2t = the second measure of SMOOTHACT, calculated as the correlation between change in discretionary accruals and change in earnings adjusted
for discretionary accruals (DA).
SMOOTHACT3t = the third measure of SMOOTHACT, calculated the same as the first, except that in measuring SMOOTHXEM 3t, it employs
performance-matched discretionary accruals, following Kothari et al. (2005). SMOOTHXEM3t is measured the same as SMOOTHXEM1t except that it
employs performance-matched discretionary accruals. Each sample firm is matched with a control firm on year, industry membership based on two-digit
SIC, and return on asset. The difference in DA between each pair of sample and control firms is performance-matched discretionary accruals.
NOAt-1 = net operating assets scaled by lagged sales, measured at the beginning of the year as [shareholders’ equity (#216) - cash and marketable securities (#1)
+ total debt (#9 + #34)] / sales (#12).
BMt = book-to-market ratio, measured as data60 / (data199*data25).
DEISSt+1= 1 if the amount of net bond and equity issuance in year t+1 is above the sample median, and 0 otherwise. Net bond and equity issuance is measured as
[((data111-data114+data301) + (data108-data115)) / data6].
LNSALESt = the logarithm of sales (data12).
ROAt = earnings before extraordinary and discontinued items divided by total assets [data123/lagged data6].
D_POSt = 1 if ROA increases from last year, and 0 otherwise. ROAt is defined as earnings before extraordinary and discontinued items divided by total assets at
the beginning of the year [data123/lagged data6].
ΔLEVt = change in long-term debt divided by total assets [data9 / data6].
ΔSALESt = change in sales revenue scaled by lagged total assets [data12 / lagged data6].
ΔFCFt = change in free cash flow scaled by lagged total assets. Free cash flow is measured as cash flow from operating activities minus average capital
expenditure over current and the past two years [(data308-average of data128) / lagged data6].
ΔPPEt = change in gross PPE divided by lagged total assets [data7 / lagged data6].
ΔSIZEt = change in natural log of total assets [data6].
ΔINT_COVt = change in operating income before depreciation divided by interest expense [data13 / data15].
39
Table 2: Regression of the Extent of Smoothing on Credit Ratings
SMOOTHACT1t
Dependent Variable
Pred.
Independent
variables
Sign
INTERCEPT
D_CHGt-1
+
DPLUSRt-1
+
DMINUSRt-1
+
D_At-1
D_BBBt-1
D_BBt-1
D_Bt-1
D_CCCt-1
BMt
LNSALESt
DEISSt+1
NOAt-1
Control for firm clustering
Industry effects
Year effects
N
Likelihood Ratio
R2
Model (1)
SMOOTHACT2t
Model (2)
Model (1)
SMOOTHACT3t
Model (2)
Model (1)
Model (2)
Coeff
p
Coeff
p
Coeff
p
Coeff
p
Coeff
p
Coeff
p
0.006
0.015
0.942
0.008
0.017
0.846
0.904
0.037
<.0001
0.054
0.880
<.0001
0.047
0.013
0.654
0.022
0.063
0.553
0.020
0.010
0.025
0.007
-0.013
-0.036
-0.047
0.003
0.014
-0.001
0.031
0.003
0.087
0.080
0.631
0.420
0.058
0.135
0.146
<.0001
0.828
<.0001
0.026
0.049
0.095
0.105
0.035
0.016
0.005
0.002
0.005
0.002
0.007
0.177
0.038
0.059
0.044
0.543
0.806
0.963
0.526
0.689
0.936
0.219
0.021
0.006
0.004
-0.031
-0.037
-0.059
-0.066
0.002
0.015
-0.003
0.028
0.003
0.228
0.814
0.059
0.044
0.005
0.050
0.281
<.0001
0.693
<.0001
0.027
0.010
-0.010
-0.029
-0.039
0.003
0.015
-0.002
0.031
0.056
0.476
0.538
0.106
0.196
0.131
<.0001
0.811
<.0001
Yes
Yes
Yes
9373
Yes
Yes
Yes
9373
46%
46%
0.090
0.098
0.028
0.001
-0.011
0.002
0.003
0.003
0.007
Yes
Yes
Yes
8542
503.20
0.071
0.056
0.625
0.990
0.916
0.532
0.783
0.927
0.213
Yes
Yes
Yes
8542
503.80
0.007
-0.026
-0.032
-0.049
-0.055
0.003
0.016
-0.003
0.028
0.658
0.107
0.078
0.015
0.096
0.250
<.0001
0.666
<.0001
Yes
Yes
Yes
9437
Yes
Yes
Yes
9437
39%
39%
Note: All p values are based on the two-tailed t-statistic except those on variables with predicted signs: D_CHGt-1, DPLUSRt-1, and DMINUSRt-1. For these variables,
p values are based on one-tailed t-statistic.
D_CHGt-1=1 if data280 equals to 4, 6, 7, 9, 10, 12, 13, 15, 16, 18, 19, or 21, and 0 otherwise. D PLUSRt-1 = 1 if data280 equals to 4, 7, 10, 13, 16 or 19, and 0
otherwise. DMINUSRt-1 = 1 if data280 equals to 6, 9, 12, 15, 18, or 21. D_At-1 = 1 for firms with a credit rating of A (including A+, A, and A-), and 0 otherwise.
D_BBBt-1, D_BBt-1. D_Bt-1. and D_CCCt-1 are defined similarly. Other variables are defined the same as in table 1.
40
Table 3: Extent of Smoothing by Financing Needs (p value in parenthesis)
Dependent variable
SMOOTHACT1t
Model (5)
Model (6)
Independent
Pred.
variables
Sign
Coeff
p
Coeff
p
Panel A: Regression results
INTERCEPT
0.936
0.018
0.830
0.007
+
D_CHGHLEVt-1
0.006
0.018
+
D_CHGLLEVt-1
0.044
0.012
+
D_PLUSRHLEVt-1
0.019
0.021
+
D_PLUSRLLEVt-1
0.021
0.007
+
D_MINUSRHLEVt-1
0.018
0.020
+
D_MINUSRLLEVt-1
0.002
0.423
D_A t-1
0.063
0.023
0.114
0.026
D_BBBt-1
0.531
0.004
0.803
0.009
D_BBt-1
0.461
-0.018
0.288
-0.012
D_Bt-1
0.082
-0.039
0.040
-0.031
D_CCCt-1
0.172
-0.049
0.114
-0.042
BM t
0.136
0.003
0.139
0.003
LNSALESt
<.0001
0.014
<.0001
0.015
DEISSt+1
0.792
-0.001
0.819
-0.002
NOAt-1
<.0001
0.030
<.0001
0.030
Control for firm clustering
Yes
Yes
Industry effects
Yes
Yes
Year effects
Yes
Yes
Panel B: p-values for tests of coefficient differences
D_CHGHLEVt-1 > D_CHGLLEVt-1
0.190
D_PLUSRHLEVt-1 > D_PLUSRLLEVt-1
0.401
D_MINUSRHLEVt-1 > D_MINUSRLLEVt-1
N
9373
Likelihood Ratio
R2
46%
SMOOTHACT2t
Model (5)
Model (6)
Coeff
p
Coeff
0.903
0.069
0.009
0.085
0.087
0.009
-0.022
-0.031
0.003
0.003
0.001
0.006
Yes
Yes
Yes
<.0001
0.008
0.378
0.090
0.093
0.870
0.730
0.766
0.440
0.772
0.984
0.282
0.883
<.0001
0.058
0.000
0.082
0.014
0.088
0.091
0.013
-0.011
-0.019
0.003
0.005
0.000
0.006
Yes
Yes
Yes
0.053
0.497
0.009
0.344
0.084
0.085
0.819
0.875
0.860
0.435
0.695
0.994
0.293
46%
0.049
0.022
0.006
0.006
-0.029
-0.037
-0.055
-0.061
0.003
0.016
-0.003
0.027
Yes
Yes
Yes
0.023
8542
506. 96
0.643
0.002
0.218
0.724
0.073
0.043
0.007
0.067
0.270
<.0001
0.618
<.0001
p
0.066
0.530
0.026
0.018
0.019
-0.007
0.000
-0.036
-0.045
-0.067
-0.073
0.002
0.014
-0.003
0.027
Yes
Yes
Yes
0.003
0.023
0.018
0.221
0.977
0.028
0.017
0.002
0.031
0.284
<.0001
0.652
<.0001
0.016
0.051
9373
p
SMOOTHACT3t
Model (5)
Model (6)
Coeff
p
Coeff
8542
507.96
0.079
0.201
0.079
0.006
9437
9437
39%
39%
Note: All p values are based on the two-tailed tests except those on variables with predicted signs. For these variables, p values are based on one-tailed tests.
41
D_CHGHLEVt-1 = 1 for firms with D_CHGt-1 at 1 and a leverage ratio above the median leverage for all firms with D_CHGt-1 at 1, and 0 otherwise. Leverage ratio
is defined at long-term debt divided by lagged total assets [data9 / data6]. D_CHG LLEVt-1 = 1 for firms with D_CHGt-1 at 1 and a leverage ratio below or at the
median leverage for all firms with D_CHGt-1 at 1, and 0 otherwise. D_PLUSRHLEVt-1 and D_PLUSRLLEVt-1 ( D_MINUSRHLEVt-1 and D_MINUSRLLEVt-1) are
defined analogously among firms with a plus (minus) notch. All other variables are as defined in tables 1 and 2.
42
Table 4: The Relation between Credit Rating Changes in the Subsequent Year and Smoothness Changes in Current Year
ΔSMOOTHACT1t
Smoothing Measure:
Variable
INTERCEPT
ΔSMOOTHt
D_POSt*ΔSMOOTHt
ΔSMOOTHXEMt
D_POSt*ΔSMOOTHXEMt
ΔSMOOTHACTt
D_POSt*ΔSMOOTHACTt
D_POSt
ΔLEVt
ΔSALESt
ΔFCFt
ΔPPEt
ΔSIZEt
ΔINT_COVt
Year Effect
Industry Effect
N
Likelihood Ratio
Pred.
sign
+
+
+
-
Model (7)
Coeff
p
1.315 <.0001
0.288
0.000
-0.445 <.0001
0.595 <.0001
-1.084 <.0001
0.286
0.000
0.387
0.094
0.105
0.232
0.679 <.0001
0.000
0.603
Yes
Yes
7006
392.10
ΔSMOOTHACT2t
Model (8)
Coeff
p
Model (8)
Coeff
0.148
0.665
1.031
1.234
-1.536
1.384
-1.720
1.999
-1.084
0.285
0.355
0.099
0.686
0.000
Yes
Yes
7006
397.24
0.000
0.000
<.0001
<.0001
<.0001
<.0001
0.000
0.124
0.260
<.0001
0.602
0.166
-0.205
0.489
-1.243
0.219
0.023
0.207
0.760
0.000
Yes
Yes
4738
258.58
ΔSMOOTHACT3t
p
<.0001
0.031
0.046
<.0001
<.0001
0.027
0.937
0.072
<.0001
0.474
Model (7)
Coeff
p
Model (8)
Coeff
P
1.301 <.0001
0.249
0.002
-0.378
0.001
0.072 <.0001
0.562 <.0001
-1.175 <.0001
0.238
0.004
0.291
0.239
0.211
0.024
0.671 <.0001
0.000
0.689
Yes
Yes
6284
338.35
1.334
0.000
-1.624
0.002
1.383
0.000
-1.716
0.001
2.041 <.0001
-1.183 <.0001
0.239
0.004
0.279
0.239
0.211
0.024
0.674 <.0001
0.000
0.689
Yes
Yes
6284
340.39
Note: All p values are based on two-tailed tests except those on variables with predicted signs. For these variables, p values are based on one-tailed tests. UPt+1
(DOWNt+1) = 1 if an issuer’s credit rating is upgraded (downgraded) in the subsequent year, and 0 otherwise. ∆SMOOTHt = the rank of change in earnings
smoothness. Earnings smoothness is measured as [-s.d.(data123 for years t-4 to t)/s.d(data308 for years t-4 to t)]. ∆SMOOTHXEMt = the rank of change in
earnings smoothness adjusted for discretionary smoothing activities. Earnings smoothness adjusted for discretionary smoothing activities is measured as
[-s.d.((data123-abnormal accruals) for years t-4 to t) / s.d(data308 for years t-4 to t)], where abnormal accruals is defined the same as in table 1.
∆SMOOTHACTt = the rank of the extent of earnings smoothing , which is defined the same as in table 1.
43
Table 5: Earnings Smoothing Activities Conditional on Three Earnings Thresholds
Variable
Pred.
Sign
Panel A
p
Model (2)
Coeff
p
Subsample with EPS decrease
D_CHGt-1
DPLUSRt-1
+
+
DMINUSRt-1
N
R2
+
0.018
0.012
4026
46%
Panel B
D_CHGt-1
DPLUSRt-1
DMINUSRt-1
N
R2
Model (1)
Coeff
0.016
Panel C
0.012
0.018
0.021
0.015
4026
46%
0.051
0.005
5126
47%
0.279
5126
47%
+
+
DMINUSRt-1
N
R2
+
0.003
2507
46%
Subsample with positive EPS
0.013
0.010
0.021
2004
51%
0.225
0.062
Subsample that misses analyst forecast
D_CHGt-1
DPLUSRt-1
0.060
0.022
0.084
2004
51%
Model (2)
Coeff
P
Subsample with EPS increase
0.011
Subsample with negative EPS
+
+
+
Model (1)
Coeff
p
0.393
0.017
0.082
-0.012
2507
46%
0.150
0.028
0.019
0.008
0.006
0.203
7369
7369
46%
46%
Subsample that meets/beats
analyst forecast
0.021
0.004
0.027
0.002
4891
45%
0.015
4891
45%
0.052
Note: Models (1) and (2) are estimated with SMOOTHACT1 as the dependent variable. For parsimony, only the
variables of interest are included. D_CHGt-1=1 if data280 equals to 4, 6, 7, 9, 10, 12, 13, 15, 16, 18, 19, or 21, and 0
otherwise. DPLUSRt-1 = 1 if data280 equals to 4, 7, 10, 13, 16 or 19, and 0 otherwise. D MINUSRt-1 = 1 if data280 equals
to 6, 9, 12, 15, 18, or 21. SMOOTHACT1 is as defined in Table 1.
44
Figure 1: Extent of Smoothing by Firms with a Plus or Minus Notch Relative to Other Firms
within the Same Broad Rating Category
Panel A: The mean extent of smoothing (SMOOTHACT1t)
0.6
0.5
0.4
0.3
0.2
0.1
CC
C
CC
C-
C+
B-
B-
CC
B
B
B+
BB
BB
BB
B
BB
BBB
+
B+
A
BB
A
+
A
A
A
A
A
A
A
+
0
Panel B: The median extent of smoothing (SMOOTHACT1t)
CC
C
CC
C-
C+
CC
BB
BB
B+
BB
B
BB
BBB
+
B+
A
-
BB
A
+
A
A
A
A
A
A
A
+
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Note:
The extent of smoothing is proxied by the rank of SMOOTHACT1t. SMOOTHACT1t = [-s.d.(data123 for years t-4
to t)/s.d(data308 for years t-4 to t)] - [-s.d.((data123-abnormal accruals) for years t-4 to t)/s.d(data308 for years t-4 to
t)] as defined in Table 1.
45
Figure 2: Standard Deviation of Earnings and Cash Flows of Firms with a Plus or Minus Notch
Relative to Those of Other Firms within the Same Broad Rating
Panel A: Mean standard deviation of earnings scaled by total assets
C
CCC
CC
C+
B-
CC
B
B+
BB
-
BB
ABB
B+
BB
B
BB
BBB
+
A
A+
AA
-
AA
AA
+
0.10
0.08
0.06
0.04
0.02
0.00
Panel B: Mean standard deviation of cash flows scaled by total assets
0.06
0.05
0.04
0.03
0.02
0.01
C
C
C
-
C
C
C
+
C
C
C
B-
B
B+
BB
BB
+
BB
B
BBB
BB
B+
BB
A-
A
A+
AA
AA
AA
+
0.00
Note:
The standard deviation of earnings is measured over years t-4 to t, earnings is deflated with total assets. This term is
the numerator of SMOOTHACT1t. The standard deviation of cash flows is measured over the same period and
deflated the same way. It is the denominator of SMOOTHACT 1t as defined in Table 1.
46
Appendix: Credit Rating Revisions across Broad Rating Categories and Those within a
Broad Rating Category: the Difference of Stock Market Reactions
This appendix investigates whether the stock market responds more strongly to rating
revisions across two adjacent broad ratings (hereafter, ACROSS revisions) than to those within
the same broad rating category (hereafter, WITHIN revisions). Evidence of stronger market
reactions to ACROSS revisions would provide support for our conjecture that firms near changes
to adjacent broad rating categories have greater incentives to improve or maintain their ratings.
Bond rating change announcements are collected from FISD. FISD contains bond rating
history from credit rating agencies since April 1995. The initial sample consists of all bond rating
changes of U.S. domestic taxable corporate bonds announced by S&P and Moody’s between
April 1995 and December 2006. Since subsidiary and subordinated debt of the same firm have
the same rating changes as senior debt, we exclude them. In addition, the following bonds are
removed: Yankee bonds, issues denominated in a foreign currency, bonds issued through private
placement, and convertible bonds. We also make the following adjustments to rating changes.
When there are multiple rating changes to bond issues of the same issuer on the same date, we
retain only the issue for which the rating change is the largest, since this issue appears to affect
the stock price the most (Jorion et al. 2005). For two consecutive rating changes within three
days for the same firm, we delete the second change. Following Jorion et al. (2005), we delete
rating changes related to ‘M&A activities’ and ‘market conditions’ from the sample since these
events could contaminate the information content of rating news. Bond rating revisions
associated with default or bankruptcy filings are also deleted as they may cause extreme stock
market reactions (Goh and Ederington 1999). We require stock return data from CRSP to
compute abnormal stock returns around bond rating change announcements. The final sample for
downgrades (upgrades) has 2,311 (1,070) observations.
47
To compare the stock market reactions to ACROSS revisions with those to WITHIN
revisions, we employ the following regressions.
CAR = α0 + α1D_ACROSS + α2CHA_RATING + α3I/SGRADE +
α4RANK_BROAD + α5LOG (DAYS) + ε
(A1)
CAR = α0 + α1D_ACROSS + α2CHA_RATING + α3I/SGRADE + α4D_AA +
α5D_A + α6D_BBB + α7D_BB + α8D_B + α9D_CCC + α10D_CC +
α11LOG (DAYS) + ε
(A2)
Where,
CAR = cumulative abnormal returns, computed over the event window (-1, 1), where day 0 is the
rating change announcement date. We define abnormal return as the daily stock return for
a firm minus the daily value-weighted NYSE/AMEX market index return on the same
day. To control for the effect of extreme values of CAR, we winsorize observations that
are in the top and bottom 1 percent of its distribution.
D_ACROSS = 1 for ACROSS revisions, and 0 otherwise. For example, if a rating is downgraded
from AA to AA- by S&P, D_ACROSS is set to 0. If a rating is downgraded from AA- to
A+ by S&P, D_ACROSS is set to 1.
CHA_RATING = the absolute magnitude of the rating change (i.e., cardinal value of new rating
– cardinal value of old rating). The cardinal values range from 1 (for the best rating) to 23
(for the poorest rating).
I/SGRADE =1 if a bond rating is downgraded (upgraded) from an investment (a non-investment)
grade to a non-investment (investment) grade, and 0 otherwise. An investment grade is
defined as at or better than a BBB rating assigned by S&P or a Baa rating assigned by
Moody’s. Prior literature provides evidence that downgrades from investment to noninvestment grade generate more negative stock returns than those within the investment
grade or non-investment grade. I/SGRADE controls for this effect.
D_AA, D_A…….… and D_C are dummy variables indicating each broad rating category. For
example, D_AA equals 1 if prior bond ratings are AA+, AA, AA- (S&P) or Aa1, Aa2,
and Aa3 (Moody’s). Jorion and Zhang (2007) show that rating levels prior to ratings
changes significantly affect stock market reactions to rating changes.
LOG (DAYS) is the natural log of the number of days since the previous rating change in the
same direction. In other words, ‘DAYS’ is computed as days between two consecutive
rating changes in the same direction (i.e., downgrade after downgrade or upgrade after
upgrade). Following Jorion et al. (2005), if two consecutive rating changes are not in the
same direction, we set ‘DAYS’ equal to ‘1200’. We do not have a predicted sign on LOG
(DAYS). If LOG (DAYS) is large, each rating change is more likely to provide new
information to the stock market. On the other hand, the larger distance can attenuate
48
informational effects of a rating change due to various confounding events between two
bond ratings (Jorion et al. 2005).
RANK_BROAD is a cardinal measure of prior bond ratings. For instance, a prior bond rating of
AAA (C) is assigned ‘1’ (‘9’).
A negative (positive) coefficient on D_ACROSS in models (A1) and (A2) for
downgrades (upgrades) supports stronger market reaction to ACROSS revisions than to WITHIN
revisions. Table A1 presents the results for upgrades and downgrades separately. The coefficient
on D_ACROSS is negative (positive) for downgrades (upgrades) in both models, indicating that
ACROSS revisions lead to stronger stock market reactions than WITHIN revisions.28 Prior
literature on stock market response to credit rating revision announcements documents a weakly
significant or insignificant market response to rating upgrades (e.g. Ederington and Goh 1998).
Consistent with the literature, results in table A1 are much weaker for upgrades than for
downgrades, both in terms of the R-squared and significance of the coefficients on the
explanatory variables. In sum, our results support a stronger stock market reaction to rating
revisions across broad rating categories than to those within a broad rating category.29
28
We also compare the cumulative abnormal returns (CAR) between rating changes across broad rating categories
and rating changes within a broad rating category to show the economic significance of our results. For downgrades,
CAR for rating changes across broad rating categories is much lower than CAR for rating changes within a broad
rating category (-4.8% vs. -2.3%). For upgrade, CAR for the former group is higher than CAR for the latter group
(0.63% vs. 0.4%).
29
As a sensitivity test, we also run models (A1) and (A2) without CHA_RATING and only for one-notch bond
rating changes (i.e., when CHA_RATING =1). The results are qualitatively similar for downgrades while the
coefficient on D_ACROSS is not significant for upgrades. The results are also robust to CAR measurement based on
the market model.
49
Table A1: The Stock Market Reaction to Bond Downgrades and Upgrades ACROSS Broad Rating Categories and WITHIN a
Broad Rating Category
Dependent Variable: CAR
DOWNGRADES
Model (A1)
Pred.
Sign
Model (A2)
p
Coeff
p
0.0691
<.0001
0.0165
0.6517
-0.0140
0.0044
-0.0135
0.0112
CHA_RATING
-0.0098
0.0002
-0.0105
I/SGRADE
-0.0032
0.6794
-0.0002
RANK_BROAD
-0.0171
<.0001
INTERCEPT
D_ACROSS
-
Coeff
UPGRADES
Model (A1)
Pred.
Sign
Model (A2)
Coeff
p
0.0027
0.7275
-0.0672
0.1080
0.0039
0.1053
0.0045
0.0840
<.0001
-0.0010
0.5699
-0.0004
0.8129
0.9819
-0.0043
0.2929
-0.0068
0.2132
0.0009
0.4281
+
Coeff
p
D_AA
0.0025
0.9453
0.0630
0.1366
D_A
0.0020
0.9546
0.0732
0.0747
D_BBB
-0.0180
0.6159
0.0705
0.0852
D_BB
-0.0188
0.5977
0.0748
0.0690
D_B
-0.0447
0.2087
0.0737
0.0718
D_CCC
-0.0893
0.0141
0.0766
0.0618
D_CC
-0.2054
0.0002
0.0546
0.2136
-0.0001
0.9166
-0.0002
0.8539
LOG(DAYS)
-0.0002
0.8921
-0.0002
0.7776
Adj. R2
5.7%
6.5%
0.0%
0.0%
N
2,311
2,311
1,070
1,070
Note:
All p values are based on two-tailed t-tests except those for variables with predicted signs.
50